Mayouran VARATHALINGAM¶

Clustering Of Mobile Price on Today's Market

In [1]:
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib.colors import LinearSegmentedColormap
In [2]:
dfMobile = pd.read_csv("Mobile_price_id.csv", sep=';', index_col=0)
In [3]:
dfMobile.head(10)
Out[3]:
battery_power bluetooth clock_speed dual_sim fc four_g int_memory m_dep mobile_wt n_cores ... px_height px_width ram sc_h sc_w talk_time three_g touch_screen wifi price_range
Index
0 842 0.0 2.2 0 1 0 7 0.6 188 2 ... 20 756 2549 9 7 19 0 0 1 1
1 1021 1.0 0.5 1 0 1 53 0.7 136 3 ... 905 1988 2631 17 3 7 1 1 0 2
2 563 1.0 0.5 1 2 1 41 0.9 145 5 ... 1263 1716 2603 11 2 9 1 1 0 2
3 615 1.0 2.5 0 0 0 10 0.8 131 6 ... 1216 1786 2769 16 8 11 1 0 0 2
4 1821 1.0 1.2 0 13 1 44 0.6 141 2 ... 1208 1212 1411 8 2 15 1 1 0 1
5 1859 0.0 NaN 1 3 0 22 0.7 164 1 ... 1004 1654 1067 17 1 10 1 0 0 1
6 1821 NaN 1.7 0 4 1 10 0.8 139 8 ... 381 1018 3220 13 8 18 1 0 1 3
7 1954 0.0 0.5 1 0 0 24 0.8 187 4 ... 512 1149 700 16 3 5 1 1 1 0
8 1445 1.0 0.5 0 0 0 53 0.7 174 7 ... 386 836 1099 17 1 20 1 0 0 0
9 509 1.0 NaN 1 2 1 9 0.1 93 5 ... 1137 1224 513 19 10 12 1 0 0 0

10 rows × 21 columns

1.Data collection and preprocessing & data transformation

In [4]:
dfMobile.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 2000 entries, 0 to 1999
Data columns (total 21 columns):
battery_power    2000 non-null int64
bluetooth        1995 non-null float64
clock_speed      1995 non-null float64
dual_sim         2000 non-null int64
fc               2000 non-null int64
four_g           2000 non-null int64
int_memory       2000 non-null int64
m_dep            2000 non-null float64
mobile_wt        2000 non-null int64
n_cores          2000 non-null int64
pc               2000 non-null int64
px_height        2000 non-null int64
px_width         2000 non-null int64
ram              2000 non-null int64
sc_h             2000 non-null int64
sc_w             2000 non-null int64
talk_time        2000 non-null int64
three_g          2000 non-null int64
touch_screen     2000 non-null int64
wifi             2000 non-null int64
price_range      2000 non-null int64
dtypes: float64(3), int64(18)
memory usage: 343.8 KB

a. Fill missing values & Data cleaning

We can see that ,there are some missing values. Here "bluetooth" and "clock_speed" columns have respectively 5 missing values. Furthermore, many columns are categorical data but are represented as float data.

So we decided to keep categorical features with numeric features

In [5]:
dfMobile["bluetooth"] = dfMobile["bluetooth"].fillna(dfMobile["bluetooth"].mode()[0])
dfMobile["clock_speed"] = dfMobile["clock_speed"].fillna(round(dfMobile["clock_speed"].mean(),2))
In [6]:
dfMobile.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 2000 entries, 0 to 1999
Data columns (total 21 columns):
battery_power    2000 non-null int64
bluetooth        2000 non-null float64
clock_speed      2000 non-null float64
dual_sim         2000 non-null int64
fc               2000 non-null int64
four_g           2000 non-null int64
int_memory       2000 non-null int64
m_dep            2000 non-null float64
mobile_wt        2000 non-null int64
n_cores          2000 non-null int64
pc               2000 non-null int64
px_height        2000 non-null int64
px_width         2000 non-null int64
ram              2000 non-null int64
sc_h             2000 non-null int64
sc_w             2000 non-null int64
talk_time        2000 non-null int64
three_g          2000 non-null int64
touch_screen     2000 non-null int64
wifi             2000 non-null int64
price_range      2000 non-null int64
dtypes: float64(3), int64(18)
memory usage: 343.8 KB

Now the dataset is processed.

2.Data analysis & Visualisation

In [7]:
plt.figure(figsize=(20,20), dpi=200)

plt.subplot(4,5,1)
sns.boxplot(x = 'battery_power', data = dfMobile)

plt.subplot(4,5,2)
sns.boxplot(x = 'bluetooth', data = dfMobile)

plt.subplot(4,5,3)
sns.boxplot(x = 'clock_speed', data = dfMobile)

plt.subplot(4,5,4)
sns.boxplot(x = 'dual_sim', data = dfMobile)

plt.subplot(4,5,5)
sns.boxplot(x = 'fc', data = dfMobile)

plt.subplot(4,5,6)
sns.boxplot(x = 'four_g', data = dfMobile)

plt.subplot(4,5,7)
sns.boxplot(x = 'int_memory', data = dfMobile)

plt.subplot(4,5,8)
sns.boxplot(x = 'm_dep', data = dfMobile)

plt.subplot(4,5,9)
sns.boxplot(x = 'mobile_wt', data = dfMobile)

plt.subplot(4,5,10)
sns.boxplot(x = 'n_cores', data = dfMobile)

plt.subplot(4,5,11)
sns.boxplot(x = 'pc', data = dfMobile)

plt.subplot(4,5,12)
sns.boxplot(x = 'px_height', data = dfMobile)

plt.subplot(4,5,13)
sns.boxplot(x = 'px_width', data = dfMobile)

plt.subplot(4,5,14)
sns.boxplot(x = 'ram', data = dfMobile)

plt.subplot(4,5,15)
sns.boxplot(x = 'sc_h', data = dfMobile)

plt.subplot(4,5,16)
sns.boxplot(x = 'sc_w', data = dfMobile)

plt.subplot(4,5,17)
sns.boxplot(x = 'talk_time', data = dfMobile)

plt.subplot(4,5,18)
sns.boxplot(x = 'three_g', data = dfMobile)

plt.subplot(4,5,19)
sns.boxplot(x = 'touch_screen', data = dfMobile)

plt.subplot(4,5,20)
sns.boxplot(x = 'wifi', data = dfMobile)
Out[7]:
<matplotlib.axes._subplots.AxesSubplot at 0x16a63e5d940>

a. Correlation Matrix

In [8]:
import matplotlib.pyplot as plt
import seaborn as sns
plt.figure(figsize=(12,8))
ax = sns.heatmap(dfMobile.corr(), cmap='Greens', annot=True)
plt.title("Correlation Matrix", fontsize=20)
plt.show()

We can see that our label is very correlated with ram column, battery power and the dimension of the mobile

In [9]:
columns = ['battery_power', 'bluetooth', 'clock_speed', 'dual_sim', 'fc', 'four_g', 'int_memory', 'm_dep', 'mobile_wt', 'n_cores', 'pc','px_height','px_width','ram','sc_h','sc_w','talk_time','three_g','touch_screen','wifi']
x = dfMobile[columns]
label = dfMobile['price_range']

b. Visualization

In [10]:
from pandas.plotting import scatter_matrix
column_num=['battery_power','clock_speed','fc','int_memory','m_dep', 'mobile_wt', 'n_cores', 'pc','px_height','px_width','ram','sc_h','sc_w','talk_time','price_range']
dfMobile.hist(column =column_num ,figsize = (17,18));
In [11]:
DfMobile_num=dfMobile[['battery_power','pc','px_height','px_width','ram','price_range']]
sns.pairplot(DfMobile_num, hue="price_range")
Out[11]:
<seaborn.axisgrid.PairGrid at 0x16a650b9978>
In [12]:
sns.displot(DfMobile_num, x='ram')
plt.title('Ram Feature', weight='bold')
Out[12]:
Text(0.5, 1.0, 'Ram Feature')
In [13]:
sns.lmplot(x='ram', y='price_range', data=DfMobile_num,hue="price_range")
plt.yticks([0, 1, 2, 3])
plt.xlabel('Ram')
plt.ylabel('Price Range')
plt.title('Ram\'s correlation to Price Range', weight='bold')
plt.show()
In [14]:
sns.boxplot(x='price_range', y='battery_power', data=DfMobile_num)
plt.xlabel('Price Range')
plt.ylabel('Battery Power')
plt.title('Battery Power\'s correlation to Price Range', weight='bold')
plt.show()
In [15]:
four_g = dfMobile['four_g'].value_counts()
plt.title('Percentage of Mobiles with 4G', weight='bold')
labels_4g = ['4G', 'No 4G']
four_g.plot.pie(autopct="%.1f%%", labels=labels_4g)
plt.show()
In [16]:
three_g = dfMobile['three_g'].value_counts()
plt.title('Percentage of Mobiles with 3G', weight='bold')
labels_3g = ['3G', 'No 3G']
three_g.plot.pie(autopct="%.1f%%", labels=labels_3g)
plt.show()
In [17]:
n_cores = dfMobile['n_cores'].value_counts()
plt.title('Number of cores in mobile phones\n\n', weight='bold')
n_cores.plot.pie(autopct="%.1f%%", radius=1.5)
plt.show()
In [18]:
sns.boxplot(x='touch_screen',y='price_range',data=dfMobile)
plt.title('touch screen correlation to price range', weight='bold')
plt.show()
In [19]:
sns.boxplot(x='dual_sim',y='price_range',data=dfMobile)
plt.title('dual sim correlation to price range', weight='bold')
plt.show()

Let's apply some unsupervised learning algorithm and see how they perform on this dataset

c. Dimensional reductions techniques: PCA

First let's try PCA to reduce our dimensionality with Kmeans

In [20]:
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
data_scaled = scaler.fit_transform(x)
data_scaled
Out[20]:
array([[-0.90259726, -0.98708341,  0.83070733, ..., -1.78686097,
        -1.00601811,  0.98609664],
       [-0.49513857,  1.01308561, -1.25649702, ...,  0.55964063,
         0.99401789, -1.01409939],
       [-1.5376865 ,  1.01308561, -1.25649702, ...,  0.55964063,
         0.99401789, -1.01409939],
       ...,
       [ 1.53077336, -0.98708341, -0.76539011, ...,  0.55964063,
         0.99401789, -1.01409939],
       [ 0.62252745, -0.98708341, -0.76539011, ...,  0.55964063,
         0.99401789,  0.98609664],
       [-1.65833069,  1.01308561,  0.58515388, ...,  0.55964063,
         0.99401789,  0.98609664]])
In [21]:
from sklearn.decomposition import PCA

nb_components = 4
pca = PCA(n_components = nb_components)
pca.fit(data_scaled)
data_pca = pca.transform(data_scaled)
In [22]:
plt.figure(figsize=(8,5))
colnames = columns
hm = sns.heatmap(pca.components_.T, cbar=True, annot=True, cmap='bwr', fmt=".2f",
annot_kws={"size":10}, yticklabels=colnames, vmax=1, vmin=-1, center=0)
plt.xlabel("pca",fontsize=13)
plt.ylabel("variables",fontsize=13)
plt.title("Correlation between variables and pca components",fontsize=18)
Out[22]:
Text(0.5, 1.0, 'Correlation between variables and pca components')
In [23]:
plt.figure(figsize=(8,6))
sns.scatterplot(data_pca[:,1], data_pca[:,3], hue=label, legend='full')
plt.xlabel('First principal component')
plt.ylabel('Second Principal Component')

C:\Users\mayou\Anaconda3\lib\site-packages\seaborn\_decorators.py:43: FutureWarning: Pass the following variables as keyword args: x, y. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.
  FutureWarning
Out[23]:
Text(0, 0.5, 'Second Principal Component')

3.Apply unsupervised learning algorithm

a. Kmeans with PCA

In [24]:
from sklearn.cluster import KMeans
import numpy as np
X=np.array(list(zip(data_pca[:,1],data_pca[:,3])))
colors= ['b','g','c','k']
markers= ['o','v','h','+']
plt.ylabel('PCA dimension')

kmeans = KMeans(n_clusters=4).fit(X)
plt.scatter(kmeans.cluster_centers_[:,0],kmeans.cluster_centers_[:,1],s=200,c='red',label='Centroids')

for i, l in enumerate(kmeans.labels_):
    plt.plot(data_pca[:,1][i],data_pca[:,3][i],color=colors[l],marker=markers[l])
plt.xlabel(' PCA dimension ')
plt.legend()
plt.show()

Elbow Method to find number of clusters, even if we need 4 clusters like in our target variable

In [25]:
ssd = []
range_n_clusters = [2, 3, 4, 5, 6, 7, 8]
for num_clusters in range_n_clusters:
    kkmeans = KMeans(n_clusters=num_clusters, max_iter=1000)
    kkmeans.fit(X)
    
    ssd.append(kkmeans.inertia_)

plt.plot(ssd)
Out[25]:
[<matplotlib.lines.Line2D at 0x16a6ce12c50>]

Internal Index : Silhouette Score

In [26]:
from sklearn.metrics import silhouette_score
range_n_clusters = [2, 3, 4, 5, 6, 7, 8]

for num_clusters in range_n_clusters:
    
    # intialise kmeans
    kameans = KMeans(n_clusters=num_clusters, max_iter=1000)
    kameans.fit(X)
    
    cluster_labels = kameans.labels_
    
    # silhouette score
    silhouette_avg = silhouette_score(X, cluster_labels)
    print("For n_clusters={0}, the silhouette score is {1}".format(num_clusters, silhouette_avg))

For n_clusters=2, the silhouette score is 0.33984636949290237
For n_clusters=3, the silhouette score is 0.3630275884730351
For n_clusters=4, the silhouette score is 0.3392835856384422
For n_clusters=5, the silhouette score is 0.32953005390567713
For n_clusters=6, the silhouette score is 0.34122292301162543
For n_clusters=7, the silhouette score is 0.33222474061129126
For n_clusters=8, the silhouette score is 0.3300991031667386
In [27]:
dfMobile['K-Means_Cluster_ID']=kmeans.labels_
In [28]:
dfMobile.head(100)
Out[28]:
battery_power bluetooth clock_speed dual_sim fc four_g int_memory m_dep mobile_wt n_cores ... px_width ram sc_h sc_w talk_time three_g touch_screen wifi price_range K-Means_Cluster_ID
Index
0 842 0.0 2.20 0 1 0 7 0.6 188 2 ... 756 2549 9 7 19 0 0 1 1 0
1 1021 1.0 0.50 1 0 1 53 0.7 136 3 ... 1988 2631 17 3 7 1 1 0 2 2
2 563 1.0 0.50 1 2 1 41 0.9 145 5 ... 1716 2603 11 2 9 1 1 0 2 2
3 615 1.0 2.50 0 0 0 10 0.8 131 6 ... 1786 2769 16 8 11 1 0 0 2 2
4 1821 1.0 1.20 0 13 1 44 0.6 141 2 ... 1212 1411 8 2 15 1 1 0 1 2
5 1859 0.0 1.52 1 3 0 22 0.7 164 1 ... 1654 1067 17 1 10 1 0 0 1 2
6 1821 0.0 1.70 0 4 1 10 0.8 139 8 ... 1018 3220 13 8 18 1 0 1 3 1
7 1954 0.0 0.50 1 0 0 24 0.8 187 4 ... 1149 700 16 3 5 1 1 1 0 3
8 1445 1.0 0.50 0 0 0 53 0.7 174 7 ... 836 1099 17 1 20 1 0 0 0 0
9 509 1.0 1.52 1 2 1 9 0.1 93 5 ... 1224 513 19 10 12 1 0 0 0 1
10 769 0.0 2.90 1 0 0 9 0.1 182 5 ... 874 3946 5 2 7 0 0 0 3 3
11 1520 1.0 2.20 0 5 1 33 0.5 177 8 ... 1005 3826 14 9 13 1 1 1 3 1
12 1815 0.0 2.80 0 2 0 33 0.6 159 4 ... 748 1482 18 0 2 1 0 0 1 0
13 803 1.0 2.10 0 7 0 17 1.0 198 4 ... 1440 2680 7 1 4 1 0 1 2 2
14 1866 0.0 0.50 0 13 1 52 0.7 185 1 ... 563 373 14 9 3 1 0 1 0 1
15 775 0.0 1.00 0 3 0 46 0.7 159 2 ... 1864 568 17 15 11 1 1 1 0 1
16 838 0.0 1.52 0 1 1 13 0.1 196 8 ... 1850 3554 10 9 19 1 0 1 3 2
17 595 0.0 0.90 1 7 1 23 0.1 121 3 ... 810 3752 10 2 18 1 1 0 3 1
18 1131 0.0 0.50 1 11 0 49 0.6 101 5 ... 878 1835 19 13 16 1 1 0 1 1
19 682 1.0 0.50 0 4 0 19 1.0 121 4 ... 1064 2337 11 1 18 0 1 1 1 3
20 772 0.0 1.10 1 12 0 39 0.8 81 7 ... 1854 2819 17 15 3 1 1 0 3 1
21 1709 1.0 2.10 0 1 0 13 1.0 156 2 ... 1385 3283 17 1 15 1 0 0 3 3
22 1949 0.0 1.52 1 4 0 47 0.3 199 4 ... 822 1433 11 5 20 0 0 1 1 0
23 1602 1.0 2.80 1 4 1 38 0.7 114 3 ... 788 1037 8 7 20 1 0 0 0 1
24 503 0.0 1.20 1 5 1 8 0.4 111 3 ... 1245 2583 11 0 12 1 0 0 1 2
25 961 1.0 1.40 1 0 1 57 0.6 114 8 ... 1434 2782 18 9 7 1 1 1 2 1
26 519 1.0 1.60 1 7 1 51 0.3 132 4 ... 645 3763 16 1 4 1 0 1 3 1
27 956 0.0 0.50 0 1 1 41 1.0 143 7 ... 1075 3286 17 8 12 1 1 0 3 1
28 1453 0.0 1.60 1 12 1 52 0.3 96 2 ... 1311 2373 10 1 10 1 1 1 2 1
29 851 0.0 0.50 0 3 0 21 0.4 200 5 ... 1263 478 12 7 10 1 0 1 0 2
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
70 1448 1.0 0.50 1 6 1 45 0.8 138 7 ... 1724 3378 13 11 2 1 1 1 3 1
71 1407 1.0 2.40 0 1 1 22 0.7 104 4 ... 1217 2192 9 7 13 1 0 1 2 2
72 605 0.0 1.00 0 8 1 9 0.1 142 3 ... 1026 392 15 6 4 1 0 0 0 1
73 1038 0.0 1.20 0 3 0 43 0.7 141 1 ... 638 3709 11 0 12 1 1 1 3 0
74 797 1.0 2.90 1 4 0 38 0.5 90 4 ... 1413 590 9 2 6 0 1 0 0 3
75 819 0.0 0.60 1 8 1 42 0.9 188 6 ... 1242 1814 5 4 13 1 0 1 1 2
76 1114 0.0 2.80 0 4 1 9 0.4 197 3 ... 1071 907 10 7 17 1 1 0 0 2
77 1234 1.0 1.60 1 1 0 33 0.6 172 1 ... 778 1449 11 7 6 0 0 1 0 0
78 1199 1.0 2.50 1 15 1 16 0.2 116 4 ... 1552 3448 17 13 10 1 0 1 3 1
79 1103 0.0 1.00 1 6 0 29 0.7 111 6 ... 1486 1837 8 7 16 0 1 1 1 3
80 1589 1.0 0.60 1 0 1 58 0.9 85 7 ... 1206 3464 19 10 6 1 1 1 3 1
81 999 0.0 2.90 1 11 1 64 0.2 199 4 ... 1616 2593 14 11 16 1 1 0 2 1
82 1510 1.0 0.90 1 2 0 45 0.9 180 5 ... 1752 3484 9 6 11 0 1 1 3 3
83 1008 0.0 0.80 0 11 0 61 1.0 114 3 ... 1122 2009 10 3 11 1 0 0 1 2
84 1127 1.0 2.90 1 5 1 57 0.8 163 1 ... 1049 2048 11 5 17 1 1 1 1 2
85 1412 1.0 2.40 0 5 0 25 0.8 96 8 ... 1894 837 15 9 4 0 0 1 1 0
86 1496 1.0 2.00 1 4 0 42 0.5 182 5 ... 741 854 16 0 7 0 1 1 0 0
87 1083 1.0 2.90 1 1 1 64 0.8 178 3 ... 1118 3210 13 3 4 1 0 0 3 2
88 668 0.0 0.50 1 0 0 3 0.1 155 5 ... 1203 2746 9 5 8 1 1 1 2 3
89 1309 0.0 1.10 1 0 0 33 0.5 100 4 ... 1402 2334 10 1 11 1 0 1 2 3
90 1724 0.0 2.00 1 2 1 57 0.5 177 3 ... 1924 2822 19 3 16 1 1 0 3 2
91 1977 1.0 2.00 1 7 1 54 1.0 171 7 ... 1242 1971 18 7 14 1 1 0 2 1
92 885 0.0 2.30 1 0 1 15 0.4 103 7 ... 802 1410 16 10 5 1 1 1 0 1
93 879 1.0 2.50 0 11 1 14 0.7 83 6 ... 1105 349 16 10 2 1 0 1 0 1
94 1322 0.0 1.70 1 6 0 7 0.8 140 3 ... 1990 1418 19 17 12 0 1 0 1 0
95 1137 1.0 1.00 0 18 0 7 1.0 196 3 ... 1179 3616 13 5 12 1 1 1 3 1
96 1355 0.0 2.30 0 10 1 23 0.2 132 5 ... 891 880 19 4 8 1 1 0 0 1
97 1665 1.0 0.50 0 3 1 60 0.2 194 6 ... 1684 1601 18 17 14 1 0 0 2 1
98 657 0.0 2.50 0 0 0 37 0.7 141 2 ... 961 1412 14 3 18 0 1 0 0 3
99 593 0.0 0.50 0 6 0 31 0.4 156 7 ... 1317 1692 6 2 7 1 0 0 1 2

100 rows × 22 columns

External Index : NMI and Adjusted Rand Score

In [29]:
from sklearn.metrics.cluster import normalized_mutual_info_score
normalized_mutual_info_score(dfMobile['price_range'], dfMobile['K-Means_Cluster_ID'])
Out[29]:
0.003092508451085772
In [30]:
from sklearn.metrics.cluster import adjusted_rand_score
adjusted_rand_score(dfMobile['price_range'], dfMobile['K-Means_Cluster_ID'])
Out[30]:
0.001231533143980699

Characteristic of Kmeans Cluster : Exploration of the Clusters found

In [31]:
plt.figure(figsize=(20,20),dpi=200)
columns_w_pr=['battery_power', 'bluetooth', 'clock_speed', 'dual_sim', 'fc', 'four_g', 'int_memory', 'm_dep', 'mobile_wt', 'n_cores', 'pc','px_height','px_width','ram','sc_h','sc_w','talk_time','three_g','touch_screen','wifi','price_range']
for i,j in zip(range(1, 22, 1),(columns_w_pr)):
    plt.subplot(7,4,i)
    sns.boxplot(x='K-Means_Cluster_ID', y=j, data=dfMobile)

thank to this plot we can see the repartion of each features in each cluster and understand how kmeans cluster worked with our data

b. Hierarchial Clustering with PCA

We used two hierachial clustering, one from spicy and one from sklearn to see the difference

In [32]:
df_data_scaled=pd.DataFrame(data_scaled,columns=['battery_power', 'bluetooth', 'clock_speed', 'dual_sim', 'fc', 'four_g', 'int_memory', 'm_dep', 'mobile_wt', 'n_cores', 'pc','px_height','px_width','ram','sc_h','sc_w','talk_time','three_g','touch_screen','wifi'])
df_data_scaled.head()
Out[32]:
battery_power bluetooth clock_speed dual_sim fc four_g int_memory m_dep mobile_wt n_cores pc px_height px_width ram sc_h sc_w talk_time three_g touch_screen wifi
0 -0.902597 -0.987083 0.830707 -1.019184 -0.762495 -1.043966 -1.380644 0.340740 1.349249 -1.101971 -1.305750 -1.408949 -1.146784 0.391703 -0.784983 0.283103 1.462493 -1.786861 -1.006018 0.986097
1 -0.495139 1.013086 -1.256497 0.981177 -0.992890 0.957886 1.155024 0.687548 -0.120059 -0.664768 -0.645989 0.585778 1.704465 0.467317 1.114266 -0.635317 -0.734267 0.559641 0.994018 -1.014099
2 -1.537686 1.013086 -1.256497 0.981177 -0.532099 0.957886 0.493546 1.381165 0.134244 0.209639 -0.645989 1.392684 1.074968 0.441498 -0.310171 -0.864922 -0.368140 0.559641 0.994018 -1.014099
3 -1.419319 1.013086 1.199038 -1.019184 -0.992890 -1.043966 -1.215274 1.034357 -0.261339 0.646842 -0.151168 1.286750 1.236971 0.594569 0.876859 0.512708 -0.002014 0.559641 -1.006018 -1.014099
4 1.325906 1.013086 -0.397060 -1.019184 2.002254 0.957886 0.658915 0.340740 0.021220 -1.101971 0.673534 1.268718 -0.091452 -0.657666 -1.022389 -0.864922 0.730240 0.559641 0.994018 -1.014099
In [33]:
import matplotlib.pyplot as plt
import seaborn as sns
plt.figure(figsize=(10,8))
ax = sns.heatmap(df_data_scaled.corr(), cmap='Greens', annot=True)
plt.title("Correlation Matrix", fontsize=20)
plt.show()

single linkage

In [34]:
from scipy.cluster.hierarchy import linkage
from scipy.cluster.hierarchy import dendrogram
from scipy.cluster.hierarchy import cut_tree
plt.figure(figsize=(10,15))
sl_mergings = linkage(df_data_scaled, method="single", metric='euclidean')
dendrogram(sl_mergings)

plt.show()

complete linkage

In [35]:
# complete linkage
plt.figure(figsize=(30,30))
cl_mergings = linkage(df_data_scaled, method="complete", metric='euclidean')
dendrogram(cl_mergings)

plt.show()
In [36]:
cl_cluster_labels = cut_tree(cl_mergings, n_clusters=4).reshape(-1, )
cl_cluster_labels
Out[36]:
array([0, 0, 0, ..., 0, 3, 2])
In [37]:
dfMobile['Hierarchical_Cluster_labels']=cl_cluster_labels

Characteristic of Hierachial Cluster : Exploration of the Clusters found

In [38]:
plt.figure(figsize=(20,20),dpi=200)
columns_w_pr=['battery_power', 'bluetooth', 'clock_speed', 'dual_sim', 'fc', 'four_g', 'int_memory', 'm_dep', 'mobile_wt', 'n_cores', 'pc','px_height','px_width','ram','sc_h','sc_w','talk_time','three_g','touch_screen','wifi','price_range']
for i,j in zip(range(1, 22, 1),(columns_w_pr)):
    plt.subplot(7,4,i)
    sns.boxplot(x='Hierarchical_Cluster_labels', y=j, data=dfMobile)
In [39]:
from sklearn.cluster import AgglomerativeClustering

cluster_h = AgglomerativeClustering(n_clusters=4, affinity='euclidean', linkage='ward')
cluster_h.fit_predict(X)

cluster_h.labels_
Out[39]:
array([0, 1, 3, ..., 3, 2, 2], dtype=int64)

Internal Index : Silhouette Score

In [40]:
from sklearn.metrics import silhouette_score
range_n_clusters = [2, 3, 4, 5, 6, 7, 8]

for num_clusters in range_n_clusters:
    
    # intialise 
    cluster_hh = AgglomerativeClustering(n_clusters=num_clusters,affinity='euclidean', linkage='ward')
    cluster_hh.fit(X)
    
    cluster_hh_labels = cluster_h.labels_
    # silhouette score
    silhouette_avg = silhouette_score(X, cluster_hh_labels)
    print("For n_clusters={0}, the silhouette score is {1}".format(num_clusters, silhouette_avg))

For n_clusters=2, the silhouette score is 0.2655857739743639
For n_clusters=3, the silhouette score is 0.2655857739743639
For n_clusters=4, the silhouette score is 0.2655857739743639
For n_clusters=5, the silhouette score is 0.2655857739743639
For n_clusters=6, the silhouette score is 0.2655857739743639
For n_clusters=7, the silhouette score is 0.2655857739743639
For n_clusters=8, the silhouette score is 0.2655857739743639

Comparaison of Kmeans and Hierarchial Clustering

In [41]:
data_pca_df = pd.DataFrame(data_pca,columns=['PC1','PC2','PC3','PC4'])
In [42]:
data_pca_df.head()
Out[42]:
PC1 PC2 PC3 PC4
0 0.293058 -2.614122 -0.583106 1.295868
1 -1.976934 0.575289 0.371078 -1.130758
2 -1.269173 0.194788 0.074594 -1.986688
3 -1.087370 -0.242464 1.628784 -0.228918
4 1.647321 1.289198 -0.363200 -1.840295
In [43]:
data_pca_df['K_Means_Cluster_ID'] = kmeans.labels_
data_pca_df['Hierarchical_Cluster_Labels'] = cl_cluster_labels
In [44]:
data_pca_df['Hierarchical_Cluster_Labels_sklearn'] = cluster_h.labels_
In [48]:
plt.figure(figsize=(12,6),dpi=200)

plt.subplot(1,3,1)
sns.scatterplot(x='PC1',y='PC2',data=data_pca_df,hue='K_Means_Cluster_ID')


plt.subplot(1,3,3)
sns.scatterplot(x='PC1',y='PC2',data=data_pca_df,hue='Hierarchical_Cluster_Labels_sklearn')
Out[48]:
<matplotlib.axes._subplots.AxesSubplot at 0x16a643e0438>
In [46]:
dfMobile['Hierarchical_Cluster_Labels_sklearn']=cluster_h.labels_
In [52]:
plt.figure(figsize=(15,10),dpi=200)

plt.subplot(1,3,1)
sns.scatterplot(x='ram',y='battery_power',data=dfMobile,hue='K-Means_Cluster_ID')


plt.subplot(1,3,2)
sns.scatterplot(x='ram',y='battery_power',data=dfMobile,hue='Hierarchical_Cluster_Labels_sklearn')

plt.subplot(1,3,3)
sns.scatterplot(x='ram',y='battery_power',data=dfMobile,hue='price_range')
Out[52]:
<matplotlib.axes._subplots.AxesSubplot at 0x16a6dbceac8>

The results are not good at all.

That seems to make sense especially because there are all unsupervised algorithm and not all our features determined the price of Mobiles.

In fact, the mobile price doesn't determine the overall quality of a phone

Our features are non-lineair so PCA can't perform well on this dataset.

Let's try another dimensionality reduction method.

Let's try TNSE which can be utilized with non lineair features

c. Kmeans with TSNE

In [47]:
from sklearn.manifold import TSNE
x_embedded = TSNE(n_components=2, init='random',perplexity=60.0,n_iter=1000,learning_rate=1.2).fit_transform(x)
In [48]:
from scipy.spatial.distance import cdist
distortions = []
inertias = []
K = range(1, 10)
 
for k in K:
    # Building and fitting the model
    kmeanModel = KMeans(n_clusters=k).fit(x_embedded)
    kmeanModel.fit(x_embedded)
 
    distortions.append(sum(np.min(cdist(x_embedded, kmeanModel.cluster_centers_,
                                        'euclidean'), axis=1)) / x_embedded.shape[0])
    inertias.append(kmeanModel.inertia_)
In [49]:
plt.figure(figsize=(10,6),dpi=200)
sns.scatterplot(x_embedded[:,0], x_embedded[:,1], hue=label, legend='full')
Out[49]:
<matplotlib.axes._subplots.AxesSubplot at 0x1cb56129c18>

Elbow Method

In [50]:
plt.plot(K, inertias, 'bx-')
plt.xlabel('Values of K')
plt.ylabel('Inertia')
plt.title('The Elbow Method using Inertia')
plt.show()
In [51]:
# X=np.array(list(zip(x_embedded[:,0],x_embedded[:,1])))
colors= ['b','g','c','k']
markers= ['o','v','h','+']
plt.ylabel('T-SNE dimension')

Xkmeans = KMeans(n_clusters=4).fit(x_embedded)
plt.scatter(Xkmeans.cluster_centers_[:,0],Xkmeans.cluster_centers_[:,1],s=200,c='red',label='Centroids')

for i, l in enumerate(Xkmeans.labels_):
    plt.plot(x_embedded[:,0][i],x_embedded[:,1][i],color=colors[l],marker=markers[l])
plt.xlabel(' T-SNE dimension ')
plt.legend()
plt.show()

External Index :Adjusted Rand Score

In [52]:
from sklearn.metrics.cluster import adjusted_rand_score
adjusted_rand_score(dfMobile['price_range'], Xkmeans.labels_)
Out[52]:
0.43269019321563834

Using tsne with keamns seems to be a better idea than pca with kmeans

Compare distance for Kmeans

In [53]:
from pyclustering.cluster.kmeans import kmeans
from pyclustering.utils.metric import distance_metric
from pyclustering.cluster.center_initializer import random_center_initializer
from pyclustering.cluster.encoder import type_encoding
from pyclustering.cluster.encoder import cluster_encoder
from sklearn.metrics.cluster import contingency_matrix
distance_measures = {'euclidean': 0, 'squared euclidean': 1, 'manhattan': 2, 'chebyshev': 3, 
                    'canberra': 5, 'chi-square': 6,}

def purity_score(y_true, y_pred):
    #contingency matrix
    confusion_matrix = contingency_matrix(y_true, y_pred)
    return np.sum(np.amax(confusion_matrix, axis=0)) / np.sum(confusion_matrix)



def dist_compare(dist_measure):
    initial_centers = random_center_initializer(x_embedded, 4, random_state=5).initialize()
    # instance pour avoir distance differente 
    instanceKm = kmeans(x_embedded, initial_centers=initial_centers, metric=distance_metric(dist_measure))
    # process cluster
    instanceKm.process()
    # cclusters et centers
    pyClusters = instanceKm.get_clusters()
    pyCenters = instanceKm.get_centers()
    # encode
    pyEncoding = instanceKm.get_cluster_encoding()
    pyEncoder = cluster_encoder(pyEncoding, pyClusters, x_embedded)
    pyLabels = pyEncoder.set_encoding(0).get_clusters()
    #print(pyLabels)
    
    return pyLabels#purity_score(y, pyLabels)

for measure, value in distance_measures.items():
    from sklearn.metrics.cluster import adjusted_rand_score
    print(measure)
    print(adjusted_rand_score(dfMobile['price_range'], dist_compare(value)))

euclidean
0.43464680327301536
squared euclidean
0.43291598836715267
manhattan
0.4285617975628693
chebyshev
0.4315372049701328
canberra
0.3066896735611582
chi-square
0.31749946445265775

Changing distance method for Kmeans

In [55]:
from scipy.spatial import distance
x_seu=distance.cdist(x_embedded,x_embedded, 'sqeuclidean')
print(x_seu)

[[  0.         191.23233639 240.57781097 ... 103.63924007 287.77230207
   75.67688427]
 [191.23233639   0.           2.99813657 ...  14.62967268 451.14992901
  180.90951005]
 [240.57781097   2.99813657   0.         ...  29.15724822 511.88008515
  212.75188353]
 ...
 [103.63924007  14.62967268  29.15724822 ...   0.         404.33942657
  100.30713749]
 [287.77230207 451.14992901 511.88008515 ... 404.33942657   0.
  632.34651025]
 [ 75.67688427 180.90951005 212.75188353 ... 100.30713749 632.34651025
    0.        ]]

Position of Centroids

In [56]:
Xkmeans_seu = KMeans(n_clusters=4).fit(x_seu)
plt.scatter(Xkmeans_seu.cluster_centers_[:,0],Xkmeans_seu.cluster_centers_[:,1],s=200,c='red',label='Centroids')
Out[56]:
<matplotlib.collections.PathCollection at 0x1cb61461710>
In [57]:
from sklearn.metrics.cluster import adjusted_rand_score
adjusted_rand_score(dfMobile['price_range'], Xkmeans_seu.labels_)
Out[57]:
0.4743270263812373

Changing the distance method before applying Kmeans improves our adjusted rand score

d. Additional Work to improve Clustering

1.Gower distance with Kmedoids

In [118]:
import gower
from sklearn_extra.cluster import KMedoids
gower_mat = gower.gower_matrix(x_seu)

km_model = KMedoids(n_clusters = 4, random_state = 0, metric = 'precomputed', method = 'pam', init =  'k-medoids++').fit(gower_mat)
In [119]:
clusters_kmedoids = km_model.labels_
In [120]:
from sklearn.metrics.cluster import adjusted_rand_score
adjusted_rand_score(dfMobile['price_range'], clusters_kmedoids)
Out[120]:
0.42935679308294133

2.K-Prototype Algorithm

In [59]:
from kmodes.kprototypes import KPrototypes
In [71]:
dfMobileKPrototypes = dfMobile[['battery_power', 'bluetooth', 'clock_speed', 'dual_sim', 'fc', 'four_g', 'int_memory', 'm_dep', 'mobile_wt', 'n_cores', 'pc','px_height','px_width','ram','sc_h','sc_w','talk_time','three_g','touch_screen','wifi']]
In [73]:
dfMobileKPrototypes["bluetooth"] = dfMobileKPrototypes["bluetooth"].astype(str)
dfMobileKPrototypes["bluetooth"] = dfMobileKPrototypes["bluetooth"].str.replace('0.0','')
dfMobileKPrototypes["bluetooth"] = dfMobileKPrototypes["bluetooth"].astype(bool)

dfMobileKPrototypes["dual_sim"] = dfMobileKPrototypes["dual_sim"].astype(str)
dfMobileKPrototypes["dual_sim"] = dfMobileKPrototypes["dual_sim"].str.replace('0','')
dfMobileKPrototypes["dual_sim"] = dfMobileKPrototypes["dual_sim"].astype(bool)

dfMobileKPrototypes["four_g"] = dfMobileKPrototypes["four_g"].astype(str)
dfMobileKPrototypes["four_g"] = dfMobileKPrototypes["four_g"].str.replace('0','')
dfMobileKPrototypes["four_g"] = dfMobileKPrototypes["four_g"].astype(bool)

dfMobileKPrototypes["three_g"] = dfMobileKPrototypes["three_g"].astype(str)
dfMobileKPrototypes["three_g"] = dfMobileKPrototypes["three_g"].str.replace('0','')
dfMobileKPrototypes["three_g"] = dfMobileKPrototypes["three_g"].astype(bool)

dfMobileKPrototypes["touch_screen"] = dfMobileKPrototypes["touch_screen"].astype(str)
dfMobileKPrototypes["touch_screen"] = dfMobileKPrototypes["touch_screen"].str.replace('0','')
dfMobileKPrototypes["touch_screen"] = dfMobileKPrototypes["touch_screen"].astype(bool)

dfMobileKPrototypes["wifi"] = dfMobileKPrototypes["wifi"].astype(str)
dfMobileKPrototypes["wifi"] = dfMobileKPrototypes["wifi"].str.replace('0','')
dfMobileKPrototypes["wifi"] = dfMobileKPrototypes["wifi"].astype(bool)

C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:1: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  """Entry point for launching an IPython kernel.
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:2: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:3: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  This is separate from the ipykernel package so we can avoid doing imports until
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:5: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  """
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:6: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:7: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  import sys
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:9: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  if __name__ == '__main__':
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:10: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  # Remove the CWD from sys.path while we load stuff.
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:11: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  # This is added back by InteractiveShellApp.init_path()
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:13: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  del sys.path[0]
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:14: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:15: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  from ipykernel import kernelapp as app
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:17: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:18: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:19: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:21: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:22: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:23: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
In [77]:
dfMobileKPrototypes['bluetooth'] = pd.Categorical(dfMobileKPrototypes.bluetooth)
dfMobileKPrototypes['dual_sim'] = pd.Categorical(dfMobileKPrototypes.dual_sim)
dfMobileKPrototypes['four_g'] = pd.Categorical(dfMobileKPrototypes.four_g)
dfMobileKPrototypes['three_g'] = pd.Categorical(dfMobileKPrototypes.three_g)
dfMobileKPrototypes['touch_screen'] = pd.Categorical(dfMobileKPrototypes.touch_screen)
dfMobileKPrototypes['wifi'] = pd.Categorical(dfMobileKPrototypes.wifi)

C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:1: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  """Entry point for launching an IPython kernel.
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:2: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:3: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  This is separate from the ipykernel package so we can avoid doing imports until
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:4: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  after removing the cwd from sys.path.
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:5: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  """
C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:6: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
In [78]:
dfMobileKPrototypes.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 2000 entries, 0 to 1999
Data columns (total 20 columns):
battery_power    2000 non-null int64
bluetooth        2000 non-null category
clock_speed      2000 non-null float64
dual_sim         2000 non-null category
fc               2000 non-null int64
four_g           2000 non-null category
int_memory       2000 non-null int64
m_dep            2000 non-null float64
mobile_wt        2000 non-null int64
n_cores          2000 non-null int64
pc               2000 non-null int64
px_height        2000 non-null int64
px_width         2000 non-null int64
ram              2000 non-null int64
sc_h             2000 non-null int64
sc_w             2000 non-null int64
talk_time        2000 non-null int64
three_g          2000 non-null category
touch_screen     2000 non-null category
wifi             2000 non-null category
dtypes: category(6), float64(2), int64(12)
memory usage: 326.7 KB
In [79]:
categorical_features_idx=[1,3,5,17,18,19]
In [80]:
k_pro_array=dfMobileKPrototypes.values
In [81]:
kproto = KPrototypes(n_clusters=4, verbose=2, max_iter=20).fit(k_pro_array, categorical=categorical_features_idx)

Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 1, iteration: 1/20, moves: 365, ncost: 1335788196.7127144
Run: 1, iteration: 2/20, moves: 247, ncost: 1286766121.4449177
Run: 1, iteration: 3/20, moves: 132, ncost: 1267481368.001697
Run: 1, iteration: 4/20, moves: 74, ncost: 1261735189.019135
Run: 1, iteration: 5/20, moves: 37, ncost: 1259904287.0425026
Run: 1, iteration: 6/20, moves: 27, ncost: 1258768888.3110297
Run: 1, iteration: 7/20, moves: 17, ncost: 1258486836.0660462
Run: 1, iteration: 8/20, moves: 6, ncost: 1258409500.4406757
Run: 1, iteration: 9/20, moves: 1, ncost: 1258404843.3486428
Run: 1, iteration: 10/20, moves: 0, ncost: 1258404843.3486428
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 2, iteration: 1/20, moves: 495, ncost: 1347022660.676937
Run: 2, iteration: 2/20, moves: 231, ncost: 1321745323.2697577
Run: 2, iteration: 3/20, moves: 147, ncost: 1295415351.381645
Run: 2, iteration: 4/20, moves: 170, ncost: 1269322520.004055
Run: 2, iteration: 5/20, moves: 110, ncost: 1261243941.646334
Run: 2, iteration: 6/20, moves: 49, ncost: 1259165260.7156408
Run: 2, iteration: 7/20, moves: 26, ncost: 1258649503.8514829
Run: 2, iteration: 8/20, moves: 15, ncost: 1258503294.5644262
Run: 2, iteration: 9/20, moves: 9, ncost: 1258403421.0886736
Run: 2, iteration: 10/20, moves: 2, ncost: 1258393897.3380303
Run: 2, iteration: 11/20, moves: 4, ncost: 1258370845.4224887
Run: 2, iteration: 12/20, moves: 0, ncost: 1258370845.4224887
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 3, iteration: 1/20, moves: 465, ncost: 1365210718.8953986
Run: 3, iteration: 2/20, moves: 202, ncost: 1337759114.0471208
Run: 3, iteration: 3/20, moves: 101, ncost: 1327819230.735592
Run: 3, iteration: 4/20, moves: 67, ncost: 1322365486.1073427
Run: 3, iteration: 5/20, moves: 94, ncost: 1312785705.8145337
Run: 3, iteration: 6/20, moves: 105, ncost: 1298153750.1779969
Run: 3, iteration: 7/20, moves: 142, ncost: 1270557034.1945703
Run: 3, iteration: 8/20, moves: 107, ncost: 1257638952.5052345
Run: 3, iteration: 9/20, moves: 83, ncost: 1250737781.0568662
Run: 3, iteration: 10/20, moves: 47, ncost: 1248765416.249072
Run: 3, iteration: 11/20, moves: 27, ncost: 1248186171.5503166
Run: 3, iteration: 12/20, moves: 12, ncost: 1248017750.8871937
Run: 3, iteration: 13/20, moves: 2, ncost: 1248009623.6393702
Run: 3, iteration: 14/20, moves: 2, ncost: 1247999784.9239435
Run: 3, iteration: 15/20, moves: 0, ncost: 1247999784.9239435
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 4, iteration: 1/20, moves: 354, ncost: 1331043990.7173636
Run: 4, iteration: 2/20, moves: 233, ncost: 1283662240.314904
Run: 4, iteration: 3/20, moves: 123, ncost: 1268515242.6091912
Run: 4, iteration: 4/20, moves: 90, ncost: 1262362737.70824
Run: 4, iteration: 5/20, moves: 34, ncost: 1261459484.4358892
Run: 4, iteration: 6/20, moves: 17, ncost: 1260993250.2814562
Run: 4, iteration: 7/20, moves: 12, ncost: 1260787735.101425
Run: 4, iteration: 8/20, moves: 7, ncost: 1260717911.4670916
Run: 4, iteration: 9/20, moves: 6, ncost: 1260612713.2721345
Run: 4, iteration: 10/20, moves: 7, ncost: 1260526881.91059
Run: 4, iteration: 11/20, moves: 3, ncost: 1260497811.8599546
Run: 4, iteration: 12/20, moves: 1, ncost: 1260494298.63101
Run: 4, iteration: 13/20, moves: 0, ncost: 1260494298.63101
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 5, iteration: 1/20, moves: 427, ncost: 1351371961.8425052
Run: 5, iteration: 2/20, moves: 195, ncost: 1318403926.9686832
Run: 5, iteration: 3/20, moves: 106, ncost: 1307524358.4127038
Run: 5, iteration: 4/20, moves: 52, ncost: 1303469344.2493446
Run: 5, iteration: 5/20, moves: 44, ncost: 1300753909.940451
Run: 5, iteration: 6/20, moves: 29, ncost: 1299738107.5775242
Run: 5, iteration: 7/20, moves: 36, ncost: 1297862043.551715
Run: 5, iteration: 8/20, moves: 68, ncost: 1289691586.4498892
Run: 5, iteration: 9/20, moves: 92, ncost: 1274452695.448373
Run: 5, iteration: 10/20, moves: 97, ncost: 1259517693.635027
Run: 5, iteration: 11/20, moves: 81, ncost: 1251392901.7264695
Run: 5, iteration: 12/20, moves: 44, ncost: 1249100649.9282687
Run: 5, iteration: 13/20, moves: 16, ncost: 1248653540.0778313
Run: 5, iteration: 14/20, moves: 24, ncost: 1247805983.4954555
Run: 5, iteration: 15/20, moves: 14, ncost: 1247519507.505109
Run: 5, iteration: 16/20, moves: 8, ncost: 1247407148.0406885
Run: 5, iteration: 17/20, moves: 5, ncost: 1247384824.2974381
Run: 5, iteration: 18/20, moves: 0, ncost: 1247384824.2974381
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 6, iteration: 1/20, moves: 539, ncost: 1336015847.1774583
Run: 6, iteration: 2/20, moves: 209, ncost: 1315672647.849708
Run: 6, iteration: 3/20, moves: 89, ncost: 1310923194.070927
Run: 6, iteration: 4/20, moves: 28, ncost: 1310281337.8034508
Run: 6, iteration: 5/20, moves: 23, ncost: 1309846486.9018633
Run: 6, iteration: 6/20, moves: 16, ncost: 1309491826.7627916
Run: 6, iteration: 7/20, moves: 33, ncost: 1308047589.1240118
Run: 6, iteration: 8/20, moves: 38, ncost: 1306627881.3469722
Run: 6, iteration: 9/20, moves: 53, ncost: 1303574579.3726957
Run: 6, iteration: 10/20, moves: 64, ncost: 1298477779.143969
Run: 6, iteration: 11/20, moves: 54, ncost: 1294483824.8532133
Run: 6, iteration: 12/20, moves: 64, ncost: 1288812389.078308
Run: 6, iteration: 13/20, moves: 65, ncost: 1282305131.9629664
Run: 6, iteration: 14/20, moves: 94, ncost: 1269472487.2032847
Run: 6, iteration: 15/20, moves: 123, ncost: 1254449833.4702075
Run: 6, iteration: 16/20, moves: 85, ncost: 1249132236.6480777
Run: 6, iteration: 17/20, moves: 32, ncost: 1248257690.7254918
Run: 6, iteration: 18/20, moves: 14, ncost: 1248046291.8838475
Run: 6, iteration: 19/20, moves: 7, ncost: 1247999784.9239435
Run: 6, iteration: 20/20, moves: 0, ncost: 1247999784.9239435
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 7, iteration: 1/20, moves: 452, ncost: 1290124169.8626802
Run: 7, iteration: 2/20, moves: 200, ncost: 1266045360.2812688
Run: 7, iteration: 3/20, moves: 65, ncost: 1262833258.7834222
Run: 7, iteration: 4/20, moves: 38, ncost: 1261738082.504289
Run: 7, iteration: 5/20, moves: 25, ncost: 1261293095.0132117
Run: 7, iteration: 6/20, moves: 16, ncost: 1261048861.4519992
Run: 7, iteration: 7/20, moves: 6, ncost: 1260953545.674998
Run: 7, iteration: 8/20, moves: 6, ncost: 1260861189.7837906
Run: 7, iteration: 9/20, moves: 4, ncost: 1260817579.1312745
Run: 7, iteration: 10/20, moves: 7, ncost: 1260705495.4245546
Run: 7, iteration: 11/20, moves: 10, ncost: 1260520275.2959063
Run: 7, iteration: 12/20, moves: 5, ncost: 1260490753.7292767
Run: 7, iteration: 13/20, moves: 0, ncost: 1260490753.7292767
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 8, iteration: 1/20, moves: 476, ncost: 1294229202.8642166
Run: 8, iteration: 2/20, moves: 207, ncost: 1263450317.7044842
Run: 8, iteration: 3/20, moves: 58, ncost: 1260571987.7885466
Run: 8, iteration: 4/20, moves: 37, ncost: 1258698946.9572828
Run: 8, iteration: 5/20, moves: 18, ncost: 1258423386.1546617
Run: 8, iteration: 6/20, moves: 4, ncost: 1258404843.3486428
Run: 8, iteration: 7/20, moves: 0, ncost: 1258404843.3486428
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 9, iteration: 1/20, moves: 366, ncost: 1279488546.7664633
Run: 9, iteration: 2/20, moves: 127, ncost: 1265214135.2260005
Run: 9, iteration: 3/20, moves: 55, ncost: 1262841584.4683874
Run: 9, iteration: 4/20, moves: 25, ncost: 1262200655.0264049
Run: 9, iteration: 5/20, moves: 14, ncost: 1262093546.084259
Run: 9, iteration: 6/20, moves: 9, ncost: 1262027573.350804
Run: 9, iteration: 7/20, moves: 4, ncost: 1261999319.125979
Run: 9, iteration: 8/20, moves: 3, ncost: 1261964852.421694
Run: 9, iteration: 9/20, moves: 2, ncost: 1261957564.1233125
Run: 9, iteration: 10/20, moves: 0, ncost: 1261957564.1233125
Init: initializing centroids
Init: initializing clusters
Starting iterations...
Run: 10, iteration: 1/20, moves: 557, ncost: 1303173461.2583773
Run: 10, iteration: 2/20, moves: 220, ncost: 1270647801.5480719
Run: 10, iteration: 3/20, moves: 115, ncost: 1260457065.8079798
Run: 10, iteration: 4/20, moves: 63, ncost: 1257866954.3915002
Run: 10, iteration: 5/20, moves: 35, ncost: 1256813111.2437422
Run: 10, iteration: 6/20, moves: 34, ncost: 1255308959.0691562
Run: 10, iteration: 7/20, moves: 29, ncost: 1253401689.5284185
Run: 10, iteration: 8/20, moves: 33, ncost: 1250753209.2096863
Run: 10, iteration: 9/20, moves: 30, ncost: 1248951508.867046
Run: 10, iteration: 10/20, moves: 24, ncost: 1248125092.2691853
Run: 10, iteration: 11/20, moves: 9, ncost: 1248019187.7005372
Run: 10, iteration: 12/20, moves: 3, ncost: 1247999784.9239435
Run: 10, iteration: 13/20, moves: 0, ncost: 1247999784.9239435
Best run was number 5
In [83]:
clusters = kproto.predict(k_pro_array, categorical=categorical_features_idx)
In [84]:
dfMobileKPrototypes['cluster'] = list(clusters)

C:\Users\mayou\Anaconda3\lib\site-packages\ipykernel_launcher.py:1: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy
  """Entry point for launching an IPython kernel.
In [85]:
dfMobileKPrototypes.head(5)
Out[85]:
battery_power bluetooth clock_speed dual_sim fc four_g int_memory m_dep mobile_wt n_cores ... px_height px_width ram sc_h sc_w talk_time three_g touch_screen wifi cluster
Index
0 842 False 2.2 False 1 False 7 0.6 188 2 ... 20 756 2549 9 7 19 False False True 3
1 1021 True 0.5 True 0 True 53 0.7 136 3 ... 905 1988 2631 17 3 7 True True False 2
2 563 True 0.5 True 2 True 41 0.9 145 5 ... 1263 1716 2603 11 2 9 True True False 2
3 615 True 2.5 False 0 False 10 0.8 131 6 ... 1216 1786 2769 16 8 11 True False False 2
4 1821 True 1.2 False 13 True 44 0.6 141 2 ... 1208 1212 1411 8 2 15 True True False 2

5 rows × 21 columns

In [86]:
from sklearn.metrics.cluster import adjusted_rand_score
adjusted_rand_score(dfMobile['price_range'], clusters)
Out[86]:
0.4666108067332902

3.Spectral Clustering with tsne

In [61]:
from sklearn.cluster import SpectralClustering
sc = SpectralClustering( affinity='rbf',  n_clusters=4).fit(x_embedded)

labels = sc.labels_

plt.scatter(x_embedded[:,0], x_embedded[:,1], c=labels)
plt.show()
In [62]:
from sklearn.metrics.cluster import adjusted_rand_score
adjusted_rand_score(dfMobile['price_range'], sc.labels_)
Out[62]:
0.2897263721763855